Physix Frontier badgePhysix Frontier
PublishChinese | EN

Physix Frontier · News Briefing Card (IT Home · Oct 16, 2026)

Microsoft launches MAI-Base-2, claims 1/4 of OpenAI's compute

KEY FACTS

  • Microsoft AI business CEO Suleyman announced at an employee meeting that the in-house model MAI-Base-2 outperforms comparable OpenAI products.
  • Suleyman said MAI-Base-2's pretraining compute is only one quarter of OpenAI's.
  • Microsoft officially released the in-house model that day, naming it MAI-Base-2.
  • Microsoft privately compared MAI-Base-2 with the Spud model powering GPT-5.6, claiming better scores on comprehension and prediction tests.
  • Suleyman said that when compared with DeepSeek, Microsoft needs only one twelfth of its compute.

KEY DATA

1/4Pretraining compute vs OpenAI
1/12Compute vs DeepSeek

PHYSIX OBSERVATION

Microsoft is benchmarking against OpenAI with less compute, aiming to prove its in-house route can drive down large-model costs. If the data holds up, compute efficiency will become the focus of the next stage of competition, and OpenAI's technology licensing relationship will grow more delicate. But the comparison details are undisclosed, and users and developers should wait for third-party reproduction before drawing conclusions.

Source: IT Home report